To improve the accuracy of transformer performance degradation assessment, this paper presents a method for transformer degradation assessment with vibration signals on the basis of the Empirical Wavelet Transform-Multiscale entropy (EWT-MSE) and k-medoids clustering algorithm. First, transformer vibration signals are decomposed into several empirical wavelet functions (EWFs) via the EWT. Second, the MSEs of EWF components are calculated to construct the vectors of the transformer vibration signals, which enables quantification of the extracted features. Third, the healthy-state vectors and fault-state vectors serve as the training set to establish the transformer degradation assessment model through the k-medoids algorithm so that the healthy-state cluster centroid and failure-state cluster centroid of the transformer can be obtained. Last, the difference between the test data and the health status cluster centroid is calculated as the performance degradation confidence of the transformer. Experiments are presented to demonstrate the effectiveness of this method compared with other methods.